I use Claude almost every day, and I still learned a few things from a post I came across this week. Most of us learn Claude by poking around. We pick a model, paste in files, write long custom instructions and hope for the best. Nobody gives you a clear map.
That’s why this breakdown from an AI professional on LinkedIn caught my eye. The author covers Claude in six “levels.” They start with the basics and end with privacy settings most people never think about. Then they add a short plan for your first 20 minutes. I’ve expanded on each point below so you can use it right away.
1. Get the desktop app (seriously)
The creator’s first tip sounds almost too simple: stop being lazy and install the Claude desktop app. You’ll probably still use the browser about 90% of the time, and the author says that’s fine.
The difference shows up the day you start using the desktop app. You can work with your folders directly instead of uploading files one by one. You’ll also use Claude Code a lot more, because it’s right there. In the original poster’s words, “There is no coming back from there.”
Try this: Point Claude at a project folder full of notes, drafts or spreadsheets and ask it to summarize what’s inside. That’s when it clicks.
2. Stop agonizing over models
You have better things to do than spend a minute picking the “best” model every time. The expert’s default is simple: Opus 5.5 on high. That’s it.
There’s also a tip here I hadn’t thought about. Don’t switch models mid-chat. According to the post, switching uses more tokens because Claude re-reads the whole conversation to get the context again. If a task really needs a different model, start a new chat instead. It’s cleaner and it saves your usage.
3. Medium or High effort, nothing else
Effort settings confuse a lot of people, so I love how simple this rule is:
- Medium: for low-effort work like quick rewrites, short answers or simple lookups.
- High: for most of your tasks.
- Extra High: mostly burns through your usage limit, so the author skips it.
This takes away a decision you’d otherwise make a dozen times a day. Two options, clear rules, done. If a Medium answer feels thin, run it again on High and move on.
4. Delete your old Claude setup
This one surprised me the most. The LinkedIn creator recommends clearing the “Instructions for Claude” box in settings and deleting any old skills you’ve added. Why? Those prompts and skills were written for older models, and they might be slowing you down now.
The replacement is very simple. Give Claude one goal and end your prompt with this line:
“Before you start, restate my goal in your own words.”
It’s a tiny addition, but it makes Claude show you what it understood before it does any work. If it gets your goal wrong, you catch it in five seconds instead of after a full draft. I think this is one of the most useful habits in the whole post.
5. Ask for HTML, not a document
This one is clever. Add two things to your prompt: “Make it an HTML page” and “for [who will read it].”
Claude will then code the image, doc or whatever format you need. The post’s author says this leaves no room for mistakes, because HTML gives Claude full control over layout and structure. Naming the reader matters too, since it shapes the tone, the detail and the vocabulary.
Example prompt: “Summarize our Q3 results. Make it an HTML page for the leadership team.” You get something visual you can share instead of a wall of text.
6. Protect what you plug in
The last level is about privacy, and lots of people skip it. The expert’s advice is to never click the thumbs up or thumbs down under an answer. Those buttons send feedback to the AI company, and according to the post, that feedback can end up in its training data.
There’s a second warning for teams. On a Team plan, once memory is on, leave it on. If an owner turns it off, every member’s memories are deleted permanently. That’s a painful thing to do by accident, so have a quick chat with whoever manages your workspace.
Your first 20 minutes with Claude
To wrap it up, the creator shared a simple order for getting set up the right way:
- Install the desktop app.
- No settings are the best.
- Connect ONE tool, not everything.
- Use Claude Code more than you do now.
- Delete the installed skills and add new ones written for the latest models.
I really like the “connect ONE tool” step. It’s tempting to hook up your email, calendar, drive and everything else on day one. But one connection you actually use teaches you more than ten you forget about. Start with the tool you open most, like your docs or your inbox, and build from there.
Why it matters: most of the friction people feel with Claude comes from setup, not from the model. Old instructions, constant model switching and maxed-out effort settings all quietly drain your results and your usage limit. Going back to a simple setup lets the newest models do what they’re good at.
If you’ve been using Claude for a while, this is a great excuse to do a quick cleanup. Check out the full LinkedIn post to see the author’s original breakdown and join the conversation in the comments.